We study some new universal aspects of diffusion in chaotic systems, especially such having very large Lyapunov coefficients on the chaotic (indecomposable, topologically transitive) component. We do this by discretizing the chaotic component on the Surface-of-Section in a (large) number $N$ of simplectically equally big cells (in the sense of equal relative invariant ergodic measure, normalized so that the total measure of the chaotic component is unity). By iterating the transition of the chaotic orbit through SOS, where $j$ counts the number of iteration (discrete time), and assuming complete lack of correlations even between consecutive crossings (which can be justified due to the very large Lyapunov exponents), we show the universal approach of the relative measure of the occupied cells, denoted by $\rho(j)$, to the asymptotic value of unity, in the following way: $\rho(j) = 1 - (1-\frac{1}{N})^j$, so that in the limit of big $N$, $N\to \infty$, we have, for $j/N$ fixed, the exponential law $\rho(j) \approx 1 - \exp (-j/N)$. This analytic result is verified numerically in a variety of specific systems: For a plane billiard (Robnik 1983, $\lambda=0.375$), for a 3-D billiard (Prosen 1997, $a=-1/5, b=-12/5$), for ergodic logistic map (tent map), for standard map ($k=400$) and for hydrogen atom in strong magnetic field ($\epsilon=-0.05$) the agreement is almost perfect (except, in the latter two systems, for some long-time deviations on very small scale), but for H\'enon-Heiles system ($E=1/6$) and for the standard map ($k=3$) the deviations are noticed although they are not very big (only about 1%). We have tested the random number generators (Press et al 1986), and confirmed that some are almost perfect (ran0 and ran3), whilst two of them (ran1 and ran2) exhibit big deviations., Comment: 16 pages of LaTeX text, three figures (1(a-c), 2 and 3) available upon request; submitted to J. Phys. A: Math. Gen. (as a Letter)